The eye that never gets tired on your weld line

The problem

At tube-line speed, sampling leaves gaps a bad seam slips through.

100% human visual inspection is impossible at real line speed. Sample-based inspection leaves gaps between samples.

• Tube lines run at speeds where 100% human visual inspection is impossible. The eye tires, and pores, misaligned welds or cutting burrs get through.
• So inspection is sampled, typically 1-5% of parts, and everything between two samples is trusted. On a line producing thousands of tubes per shift, that is a lot of trust.
• Eddy current testing on the seam catches many weld discontinuities, but it does not see a burr from the cut-off or a surface defect the OEM rejects on sight.
• A defect found at the customer means a lot containment, sorting and a complaint on your scorecard. On steering or drive shaft tube, it can also mean a field risk.

How it fits the IRIS system

Edge — a local CNN over the line, nothing sent to the cloud.

Edge — local CNN vision, nothing sent to the cloud. A model trained on the program's specific weld and cut.

The model is trained on the weld and the cut of your specific program, not on a generic catalog, with the lighting and speed of that station. It decides in milliseconds, at line speed, and the defective tube leaves the flow before sizing, cut-to-length or bending adds value to it. Inference runs on the line itself, so it does not depend on the plant network or on the cloud. Images of your parts never leave the plant.

See the full IRIS architecture →

Before and after

Sampled inspection versus Edge on every tube

Aspect	Today	With iLEAN Edge
Share of tubes inspected	1-5% sample	100%
Pores and weld misalignment	Caught only if in the sample	Classified tube by tube
Burr from the cut-off saw	Outside eddy current's reach	Seen by the camera
Where the bad tube goes	On to the next process	Ejected at the station
Lot containments	Triggered by the customer	Sharply reduced
Images of your tubes	—	Processed locally, never sent out

1-5% sampling → 100% part inspection. Lot containments from undetected defects drop sharply.

Impact estimate

Estimated impact — to validate with your own numbers.

The block below is an estimate to be validated against your plant's actual data. We put it forward so the committee has an order of magnitude; we refine it during the assessment.

• Estimated payback 5-12 months, depending on your current reject rate and how often defects reach the customer today.
• From a 1-5% sample to 100% of tubes inspected, at real line speed and without adding an inspector.
• Lot containments from undetected defects drop sharply, together with the sorting, freight and overtime they drag along.
• Fewer OEM complaints, which is what keeps a supplier's scorecard clean for the next program award. A clean record is also what an OEM checks before sourcing the next platform.

estimated payback 5-12 months depending on current reject rate, fewer containments and OEM complaints. *Estimate to validate*.

And the fair question from the production manager

“What if it ejects good tubes?” — a false reject is the real cost of any vision system, so the model is trained on good and defective seams of your own programs, under that station's lighting. Classifying a known defect family on a fixed view is an anchored task, where the best models drop below 1.5% error [1], and borderline tubes go to a person instead of straight to scrap. Every human decision on a borderline tube feeds back into training.

^[1] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.